Method and system for predicting energy demand amount of laying hens in different day ages and growth periods through neural network model

Through neural network model, the energy demand of Beijing powder laying hens is predicted in different age groups is solved, and the problem of inaccurate energy demand prediction in the existing technology is achieved, and the energy demand of Beijing powder laying hens is accurately predicted, providing laying hen production. Theoretical basis for real-time nutritional formula.

CN118966467APending Publication Date: 2024-11-15CHINA AGRI UNIV

Patent Information

Application Number
CN202411432423.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-10-14
Publication Date
2024-11-15

AI Technical Summary

Technical Problem

The existing dynamic energy demand model is not accurate enough for the growth period energy demand prediction of Jingfen laying hens, making it difficult to achieve real-time and accurate nutritional formula design in laying hen production.

Method used

Using the neural network model, by inputting daily age and daily total energy intake, we predict the needs of metabolic energy, net energy, maintain metabolic energy, maintain net energy, deposited net energy, deposited net energy, and deposited net energy of fat. The energy data is obtained by using the respiration heat measurement method and the full fecal collection method, and a neural network model is constructed in combination with JMP software. The S-shaped TanH function or radial Gaussian function is used, with the number of hidden layers nodes of 3-5 and the learning rate is 0.1. The training and verify.

Benefits of technology

It improves the prediction accuracy of energy demand, provides the theoretical basis for realizing dynamic energy demand and real-time nutritional formula, and ensures the accuracy and reliability of prediction results.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118966467A_ABST
    Figure CN118966467A_ABST
Patent Text Reader

Abstract

The invention discloses a method for predicting the energy demand amount of laying hens at different day ages and growth periods by using a neural network model and application. The method comprises the following steps: (1) obtaining daily total energy intake of laying hens of different day ages in a growing period and corresponding energy demand data, establishing a corresponding relationship, and constructing a database; (2) setting day age and daily total energy intake as input variables and energy demand as output variables, and establishing a neural network model; and (3) taking a part of data in the database as a test set and the other part as a verification set, verifying the result of the trained neural network model, determining an available neural network model according to R2 and RMSE, and performing verification. According to the method, the day age and the daily total energy intake are input into the optimal neural network model, the metabolic energy and the net energy demand of the laying hens in the growing period are accurately predicted, a real-time and accurate energy formula can be prepared, the feed cost is saved, and the maximum production benefit is achieved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to the technical field of Jingfen laying hen breeding, and specifically designs a method and application of predicting the energy requirement of Jingfen laying hen during the growth period using a neural network method. Background Art

[0002] The energy system used in laying hen production is mainly metabolic energy. In recent years, the net energy system has attracted wide attention due to its ability to save protein feed costs and more accurately describe the effective energy value. Net energy can be divided into deposited net energy and maintenance net energy through factor analysis. Deposited net energy is the energy actually used for growth and production, and maintenance net energy is the energy for maintaining vital signs.

[0003] Energy requirements change continuously with the weight and physiological condition of growing laying hens. This makes it difficult to accurately assess real-time energy requirements in production applications. At present, there is a lack of dynamic energy requirement models for growing laying hens, so it is particularly important to monitor energy requirements using easily available indicators such as age and daily total energy intake.

[0004] Generally speaking, the method for establishing energy requirements in animal husbandry is generally linear or nonlinear regression, and the prediction accuracy is limited. In recent years, machine learning models, such as neural network models, have been increasingly used in predictive modeling between uncertain or complex variables. They do not require a functional form to describe the response surface, and their prediction accuracy is generally higher than traditional methods.

[0005] Therefore, it is urgent to establish a neural network model to predict the metabolic energy and net energy requirements of Jingfen laying hens at different ages. Summary of the invention

[0006] The present invention provides a method for predicting the energy requirement of Jingfen laying hens at different ages during growth periods. Using this method, by inputting the variable age of Jingfen laying hens during growth period and the daily total energy intake into the established optimal neural network model, the metabolic energy, net energy, maintenance metabolic energy, maintenance net energy, deposited net energy, deposited protein net energy and deposited fat net energy requirements are accurately predicted. The prediction model solves the problem of inaccurate prediction of traditional prediction models, promotes the realization of dynamic energy requirements, and provides a theoretical basis for the preparation of real-time and accurate nutritional formulas.

[0007] The present invention provides a method for establishing a prediction system for the energy requirements of Jingfen laying hens at different growth stages based on a neural network model, which comprises the following steps: (1) Obtain the daily total energy intake of laying hens of different ages and the corresponding energy requirement data, establish the corresponding relationship, and construct a database; (2) Setting age and daily total energy intake as input variables and energy requirement as output variable, a neural network model was established; (3) Part of the data in the database is a test set, and the other part is a validation set to verify the results of the trained neural network model. 2 The available neural network model was determined by RMSE and verified.

[0008] Specifically, the laying hens are Jingfen laying hens, and the energy requirements include metabolic energy, net energy, maintenance metabolic energy, maintenance net energy, deposited net energy, deposited protein net energy and deposited fat net energy requirements.

[0009] In a specific embodiment, the energy requirement is obtained by using an open respiratory calorimeter combined with a full feces collection method, and the specific determination method is: Heat production was measured by respiratory calorimetry, with 2-10 chickens in each respiratory calorimetry chamber. The appropriate number of chickens was determined according to their body weight to obtain a more accurate heat production. The chickens were allowed to adapt to the respiratory calorimetry chamber for 1 day, and the total heat production was measured for 3 days, and the fasting heat production was measured for 2 days. During the heat production measurement, the O2 consumption and CO2 emission of each respiratory calorimetry chamber were recorded, and then the heat production and fasting metabolic heat production of each chicken were calculated. The calculation formula is as follows: Heat production = 16.18 × O2 consumption + 5.02 × CO2 emissions, where the unit of heat production is kJ / d, the unit of O2 consumption is L / d, and the unit of CO2 emissions is L / d; Metabolizable energy requirement = total energy intake - fecal energy excretion, where the unit of metabolic energy requirement is kJ / d; Net energy requirement = metabolic energy requirement - (total heat production - fasting metabolic heat production), where the unit of net energy requirement is kJ / d; Maintenance metabolic energy requirement = total heat production, where the unit is kJ / d; Maintenance net energy requirement = fasting metabolic heat production, where the unit is kJ / d; Net energy requirement for deposition = metabolic energy requirement - total heat production, where the unit is kJ / d; Net energy requirement of deposited protein = (daily nitrogen intake - daily nitrogen excretion) × 6.25 × 23.86 kJ / g, where the unit of net energy requirement of deposited protein is kJ / d, and the units of daily nitrogen intake and daily nitrogen excretion are g / d; Deposited fat requirement = deposited net energy requirement - deposited protein net energy requirement, where the unit is kJ / d; Among them, the O2 consumption, CO2 emission, fasting metabolic heat production and total heat production are determined by respiratory calorimetry test; fecal energy excretion and nitrogen excretion are obtained through full fecal collection test; energy is determined by oxygen bomb calorimeter, and nitrogen is determined by Kjeldahl nitrogen analyzer.

[0010] Preferably, in step (2), a neural network model is constructed using JMP software, which adopts an S-shaped TanH function, an identity linear function or a radial Gaussian function, and the number of hidden point nodes is 3-5.

[0011] In a preferred embodiment, the algorithm involved in the neural network model is a feedforward algorithm; multi-layer perception; the hidden layer is 1 layer; the learning rate is 0.1, the number of models is 0, the penalty method is square, and the number of processes is 1.

[0012] Specifically, the data is randomly divided into a training set and a validation set, which are used to estimate model parameters and independently verify the performance of the fitted model, respectively; the validation set uses 20-40% of the data for cross-validation; the evaluation indicators of the test set and validation set are obtained: R 2 and RMSE, where R 2 The higher it is, the lower the RMSE is, and the better the model is.

[0013] The present invention provides a prediction system for energy requirements of laying hens at different growth stages based on a neural network model obtained by the establishment method.

[0014] Specifically, it includes a data input module for inputting the age of laying hens in the growing period to be predicted; a data processing module for executing the neural network model for calculation; and a result output module for outputting the data calculated by the neural network model.

[0015] The present invention also provides a method for predicting the energy requirements of laying hens at different growth stages based on a neural network model, which comprises using the prediction system to perform prediction and obtain results.

[0016] Specifically, the first step is data input, the second step is system calculation, and the third step is output of the results.

[0017] Among them, the present invention determines the input variables and the input variables based on the following analysis and considerations. In poultry science, the metabolic energy requirement is generally determined by the total feces collection method, and the net energy requirement is generally determined by the respiratory calorimetry method or the comparative slaughter method, and the determination procedure is relatively cumbersome. The energy requirement can be divided into the maintenance energy requirement and the deposition energy requirement by the factor analysis method, and the deposition energy requirement can also further include the deposition protein and deposition fat energy requirement. Specifically, maintenance energy refers to the energy requirement for maintaining vital signs, and deposition energy is the energy requirement for growth, including protein and fat growth. The metabolic energy and net energy requirements of laying hens are both derived from the total energy requirement. In addition, laying hens at different growth stages and ages have different energy requirements due to their different physiological stages and body weights, so the energy requirement is closely related to the age and total energy intake. And the age and total energy intake are easy to obtain. Therefore, in the present invention, age and daily total energy requirement are input into the neural network model as input variables, and metabolizable energy intake, net energy intake, maintenance metabolizable energy, maintenance net energy, deposited net energy, deposited protein net energy and deposited fat net energy requirements are output variables of the neural network model. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 A flow chart of a method for predicting the energy requirement of Jingfen laying hens at different growth stages provided by the present invention.

[0019] Figure 2 Schematic diagram of the optimal neural network model for predicting energy requirements.

[0020] Figure 3 Comparison between the predicted values ​​of the neural network model and the experimental values. DETAILED DESCRIPTION

[0021] The following examples are used to illustrate the present invention but are not used to limit the scope of the present invention. Example

[0022] A method for establishing a neural network to predict the energy requirements of Jingfen laying hens at different growth stages, such as Figure 1 As shown, the steps are as follows: (1) The body weight, total daily energy intake and corresponding energy requirements of Jingfen laying hens at different ages and growth stages were obtained by respiratory calorimetry and total feces collection methods: metabolizable energy, net energy, maintenance metabolic energy, maintenance net energy, deposited net energy, deposited protein net energy and deposited fat net energy requirements, and the corresponding relationships were established to construct a database.

[0023] The specific method is as follows: The animal experiment was conducted at the Animal Husbandry Branch of Jilin Academy of Agricultural Sciences. 500 0-day-old Jingfen No. 6 laying hens were selected for the experiment, and they were fed according to the feeding standards and the environmental parameters were strictly controlled. The experimental period was 0-18 weeks, and the feeding was divided into four stages: 0-6 weeks, brooding period; 7-10 weeks, early rearing period; 11-14 weeks, late rearing period; 15-18 weeks, expected production period. Corn-soybean meal standard feed with different nutritional standards was prepared according to different growth stages, as shown in Table 1.

[0024] Table 1 Composition and nutritional content of experimental diets (%, feeding basis)

[0025] An open-breathing calorimeter for chickens was used, and 16 indirect calorimetry tests and full manure collection experiments were carried out simultaneously during the trial, which lasted until 18 weeks.

[0026] Starting from 14 days of age, chickens of standard weight were selected every week for respiratory calorimetry experiments. There were 6 respiratory calorimetry chambers in total, corresponding to 6 replicates. The number of chickens in each respiratory calorimetry chamber was determined according to the chicken weight, and 2-10 chickens were placed in each respiratory calorimetry chamber.

[0027] After the chickens were placed in the respiration thermometry chamber, a week of respiration thermometry and feces collection tests were conducted. At the end of the animal experiment (18 weeks), the body weight of all Jingfen laying hens was 1328.96 ± 27.794 g.

[0028] During the formal test period, each group of chickens was placed in the respiratory thermometer chamber for one week, including 1 day of adaptation to the respiratory thermometer chamber, 3 days of formal respiratory thermometer, and 2 days of fasting thermometer. O2 consumption and CO2 emissions were recorded, and the average daily heat production was calculated.

[0029] During the formal respirometry period, a total fecal collection test was performed, in which three days of feces were mixed, dried and pulverized.

[0030] The amount of residual feed and body weight were recorded when the animals were put into and taken out of the respiration calorimeter chamber, thereby obtaining the daily weight gain and daily feed intake.

[0031] After the experiment, the nitrogen content and gross energy in the feed and feces samples were determined. The total heat production and fasting metabolic heat production were calculated by O2 consumption and CO2 emission. The total heat production is approximately considered to be the maintenance metabolic energy requirement, and the fasting metabolic heat production is approximately considered to be the maintenance net energy requirement. The metabolic energy intake is calculated by gross energy and fecal energy. The difference between the metabolic energy intake and the heat gain (total heat production minus the fasting metabolic heat production) is the net energy requirement, and the difference between the metabolic energy intake and the total heat production is the deposited net energy requirement. The net energy requirement for deposited protein is calculated by the nitrogen content. The difference between the deposited net energy and the deposited protein net energy is the deposited fat net energy.

[0032] Nitrogen content was determined by Kjeldahl method (Foss KT200, Foss, Holbeck, Denmark) with reference to GB / T6432-1994, the national standard of the People's Republic of China. Gross energy of feed and feces was determined by oxygen bomb calorimeter (IKA-C3000, IKA Group, Staufen, Germany) with reference to ISO9831:1998.

[0033] The calculation formulas for the input variables and output variables, namely, metabolic energy, net energy, maintenance metabolic energy, maintenance net energy, deposited net energy, deposited protein net energy, and deposited fat net energy, are as follows: Metabolizable energy requirement (kJ / d) = total energy intake (kJ / d) - fecal energy excretion (kJ / d) Net energy requirement (kJ / d) = metabolic energy requirement (kJ / d) - (total heat production, kJ / d - fasting metabolic heat production, kJ / d) Maintenance net energy requirement (kJ / d) = fasting metabolic heat production (kJ / d) Maintenance metabolic energy requirement (kJ / d) = total heat production (kJ / d) Net energy requirement for deposition (kJ / d) = metabolic energy requirement (kJ / d) - total heat production (kJ / d) Net energy requirement of protein deposition (kJ / d) = (daily nitrogen intake, g / d - daily nitrogen excretion, g / d) × 6.25 × 23.86 (kJ / g) Net energy requirement of fat deposition (kJ / d) = Net energy requirement of deposition (kJ / d) - Net energy requirement of protein deposition (kJ / d) The basic data required for the above calculations are as follows: Total daily heat production (kJ / d) = 16.18 × O2 consumption (L / d) + 5.02 × CO2 emissions (L / d) Daily fasting heat production (kJ / d) = average fasting heat production during fasting (kJ / d) Total energy intake (kJ / d) = feed energy content (kJ / kg) × daily feed intake (kg / d) Fecal energy excretion (kJ / d) = fecal energy content (kJ / kg) × fecal excretion (kg / d) Daily nitrogen intake (g / d) = feed nitrogen content (%) × daily feed intake (kg / d) Daily nitrogen excretion (g / d) = fecal nitrogen content (%) × fecal excretion (kg / d) (2) The optimal neural network model was established to predict the metabolic energy, net energy, maintenance metabolic energy, maintenance net energy, deposited net energy, deposited protein net energy and deposited fat net energy requirements of Jingfen laying hens at different growth stages.

[0034] Determination of parameters of the neural network model: The design software JMP pro 17 was used. The design of neural network prediction model parameters mainly includes input variables, output variables, number of hidden layers, number of hidden layer nodes, type of activation function, learning rate, number of models, penalty method, number of processes and verification method, etc. These parameters are closely related to the prediction accuracy.

[0035] First, determine the number of hidden layers. Generally, 1 layer is enough. Then determine the number of nodes. Generally, the number of nodes starts from 3 and increases continuously. In theory, the more layers and nodes there are, the better the fitting effect will be. However, the greater the amount of calculation, the more likely it is to cause overfitting. Therefore, it is necessary to design according to the specific situation.

[0036] The smaller the learning rate, the higher the number of iterations required for fitting convergence; the larger the learning rate, the less likely it is to converge.

[0037] The activation functions include three S-shaped TanH functions, the identity linear function, and the radial Gaussian function, which are designed to help the network learn complex patterns in the data, which ultimately determine what is sent to the next node. These functions are used to generate accurate sample data, and this polynomial model is not required in the neural network.

[0038] The present invention controls the input variables, output variables, activation functions, number of hidden layers and number of nodes, and the rest are default values: the verification method is "retain", the retention ratio is 33.3%, the learning rate is 0.1, the number of models is 0, the penalty method is square, and the number of processes is 1.

[0039] Determination of the Optimal Neural Network Model for Predicting Energy Requirements R 2 and RMSE represent the coefficient of determination and the root mean square error, respectively, which are indicators for measuring the accuracy of the prediction model. 2 is the coefficient of determination, ranging from 0 to 1, R 2 The larger the value, the better the fit of the measurement model to the data. RMSE stands for root mean square error. The smaller the value, the higher the prediction accuracy of the measurement model to the data. 2 The higher it is, the smaller the RMSE is, which is considered to be the better the prediction effect of the prediction model, and it is considered to be the best neural network prediction model in the present invention.

[0040] In addition, when designing the number of nodes, if R 2 Since it has become stable, there is no need to increase the number of nodes, which effectively prevents the increase in the amount of learning and training calculations in the later stage.

[0041] The specific method is as follows: The age and total energy intake of Jingfen laying hens in the growing period were matched with the energy requirements, a total of 89 sets of data were obtained, and the corresponding relationship and database were established. SPSS (International Business Machines Corporation (IBM), New York, USA) was used to analyze all the data in the database. Specifically, the data of age, daily total energy intake, metabolizable energy, net energy, maintenance metabolizable energy, maintenance net energy, deposited net energy, deposited protein net energy and deposited fat net energy requirements were sorted, and the maximum, minimum, mean and median values ​​were input by selecting the compare average value option to obtain statistical information, as shown in Table 2.

[0042] Table 2 Statistical information of the neural network basic database (kJ / d)

[0043] The neural network model was established using JMP pro 17.0 software. Age and total daily energy intake were used as input variables, and metabolizable energy, net energy, maintenance metabolizable energy, maintenance net energy, deposited net energy, deposited protein net energy, and deposited fat net energy requirements were used as output variables. 89 groups of sample data were input into the neural network model. After selecting the input variables (X, factor) and output variables (Y, response), the 89 groups of sample data were subjected to learning and training analysis. The "retained" cross-validation method was used, in which 33.3% of the samples were randomly used as cross-validation for the validation set. The activation function was selected as the S-shaped TanH function, the identity linear function, and the radial Gaussian function; the hidden layer was 1 layer; the number of hidden layer nodes was selected in the range of 3-5. The rest of the parameters used the default values: feedforward algorithm, learning rate was 0.1, number of models was 0, penalty method was square, and number of processes was 1.

[0044] Tables 3, 4, 5, 6, 7, 8 and 9 show the results of different neural network models predicting the requirements of metabolizable energy, net energy, maintenance net energy, maintenance metabolizable energy, deposited net energy, deposited protein net energy and deposited fat net energy, respectively.

[0045] Table 3 Metabolic energy requirement prediction models with different activation functions and node numbers

[0046] Table 4 Net energy requirement prediction model for different activation functions and node numbers

[0047] Table 5 Prediction model of maintenance metabolic energy requirement with different activation functions and node numbers

[0048] Table 6 Prediction model of net energy requirement for maintenance with different activation functions and node numbers

[0049] Table 7 Prediction model of net energy requirement for different activation functions and node numbers

[0050] Table 8 Prediction model of net energy of deposited protein with different activation functions and node numbers

[0051] Table 9 Prediction model of net energy of deposited fat with different activation functions and node numbers

[0052] According to R 2 The larger the value, the smaller the RMSE value, the better the number of nodes and the best function. Taking the deposition net energy as an example, the schematic diagram of the best neural network model is as follows: Figure 2 .

[0053] Results: Among the metabolic energy requirement prediction models with different activation functions and node numbers, the best activation function was the S-shaped TanH function, and the best node number was 3. 2 is 0.99, RMSE is 12.37, and the validation set R 2 The value of the activation function is 0.99, and the RMSE is 15.99. In the net energy requirement prediction model with different activation functions and node numbers, the best activation function is the radial Gaussian function, the best node number is 5, and the test set R 2 is 0.97, RMSE is 22.15, and the validation set R 2 The optimal activation function is the radial Gaussian function, the optimal number of nodes is 3, and the test set R 2 is 0.96, RMSE is 28.45, and the validation set R 2 The value of the maintenance net energy prediction model is 0.96 and the RMSE is 28.55. The best activation function is the radial Gaussian function and the best number of nodes is 4. 2 is 0.96, RMSE is 23.14, and the validation set R 2 The best activation function is the radial Gaussian function, the best node number is 4, and the test set R 2 is 0.84, RMSE is 29.84, and the validation set R 2 The best activation function is the S-shaped TanH function, the best number of nodes is 5, and the test set R 2is 0.81, RMSE is 12.98, and the validation set R 2 is 0.87, and RMSE is 10.05. In the prediction model of net energy of deposited fat with different activation functions and node numbers, R 2 None of them reached above 0.80; the best activation function was the radial Gaussian function, the best number of nodes was 4, and the test set R 2 is 0.77, RMSE is 29.40, and the validation set R 2 The average value of the regression line is 0.71 and the RMSE is 32.93.

[0054] Among the above 7 prediction models, the test set and validation set R 2 The values ​​are all greater than 0.70, and the RMSE is less than 35. This indicates that the error between the predicted value and the measured value is small, and the prediction model is relatively accurate. Figure 3 .

[0055] In order to verify the actual application of the neural network model, four new groups of data were randomly selected and verified on 33, 76, 101 and 108-day-old Jingfen laying hens. The optimal value predicted by the model was compared with the verification value (Table 10).

[0056] Table 10 Comparison of the predicted values ​​of the neural network model and the verified values ​​of energy requirements (kJ / d)

[0057] The random validation results show that the gap between the predicted value and the validation value is small (the relative error is less than 5%), indicating that the neural network model has a good prediction effect.

[0058] In summary, the present invention provides a method for predicting the metabolic energy, net energy, maintenance metabolic energy, maintenance net energy, deposited net energy, deposited protein net energy and deposited fat net energy of Jingfen laying hens at different ages in the growth period by using a neural network, and the energy requirement can be predicted by relatively simple data (age and total energy intake). The user does not need to understand the complex calculation and derivation process in the neural network, but only needs to understand its working principle, statistical principles and evaluation criteria. Therefore, the neural network model, as a simple operation and relatively accurate high-dimensional data analysis method, can effectively fit the energy requirement of Jingfen laying hens in the growth period.

Claims

1. A method for establishing a prediction system for the energy requirements of laying hens at different ages based on a neural network model, characterized in that: The steps include: (1) Obtain the daily total energy intake of laying hens of different ages and the corresponding energy requirement data, establish the corresponding relationship, and construct a database; (2) Setting age and daily total energy intake as input variables and energy requirement as output variable, a neural network model was established; (3) Part of the data in the database is a test set, and the other part is a validation set to verify the results of the trained neural network model. 2 The available neural network model was determined by RMSE and verified.

2. The method according to claim 1, characterized in that: The laying hens are Jingfen laying hens, and the energy requirements include metabolic energy, net energy, maintenance metabolic energy, maintenance net energy, deposited net energy, deposited protein net energy and deposited fat net energy requirements.

3. The method according to claim 2, characterized in that: The energy requirement is obtained by using an open respiratory calorimeter combined with a full feces collection method. The specific determination method is: Heat production was measured by respiratory calorimetry, with 2-10 chickens in each respiratory calorimetry chamber. The number of chickens was determined based on body weight to obtain accurate heat production. The chickens were allowed to adapt to the respiratory calorimetry chamber environment for 1 day, and total heat production was measured for 3 days, and fasting heat production for 2 days. During the heat production measurement, the O2 consumption and CO2 emissions of each respiratory calorimetry chamber were recorded, and then the heat production and fasting metabolic heat production of each chicken were calculated. The calculation formula is as follows: Heat production = 16.18 × O2 consumption + 5.02 × CO2 emissions, where the unit of heat production is kJ / d, the unit of O2 consumption is L / d, and the unit of CO2 emissions is L / d; Metabolizable energy requirement = total energy intake - fecal energy excretion, where the unit of metabolic energy requirement is kJ / d; Net energy requirement = metabolic energy requirement - (total heat production - fasting metabolic heat production), where the unit of net energy requirement is kJ / d; Maintenance metabolic energy requirement = total heat production, where the unit is kJ / d; Maintenance net energy requirement = fasting metabolic heat production, where the unit is kJ / d; Net energy requirement for deposition = metabolic energy requirement - total heat production, where the unit is kJ / d; Net energy requirement of deposited protein = (daily nitrogen intake - daily nitrogen excretion) × 6.25 × 23.86 kJ / g, where the unit of net energy requirement of deposited protein is kJ / d, and the units of daily nitrogen intake and daily nitrogen excretion are g / d; Deposited fat requirement = Deposited net energy requirement - Deposited protein net energy requirement, where the unit is kJ / d; Among them, the O2 consumption, CO2 emission, fasting metabolic heat production and total heat production are determined by respiratory calorimetry test; fecal energy excretion and nitrogen excretion are obtained through full fecal collection test; energy is determined by oxygen bomb calorimeter, and nitrogen is determined by Kjeldahl nitrogen analyzer.

4. The method according to claim 1, characterized in that ,In step (2), the neural network model is constructed using JMP software, which adopts the S-shaped TanH function, the identity ,linear function or radial Gaussian function, and the number of hidden point nodes is 3-5.

5. The method according to claim 4, characterized in that: The algorithm involved in the neural network model is a feedforward algorithm; multi-layer perception; 1 hidden layer; a learning rate of 0.1, a model number of 0, a penalty method of square, and a process number of 1.

6. The method according to claim 5, characterized in that: The data was randomly divided into a training set and a validation set, which were used to estimate model parameters and independently verify the performance of the fitted model, respectively. The validation set used 20-40% of the data for cross-validation. The evaluation indicators of the test set and validation set were obtained: R 2 and RMSE, where R 2 The higher it is, the lower the RMSE is, and the better the model is.

7. A prediction system for the energy requirements of laying hens at different growth stages based on a neural network model obtained according to the establishment method described in any one of claims 1 to 6.

8. The prediction system according to claim 7, characterized in that It includes a data input module for inputting the age of laying hens in the growing period to be predicted; a data processing module for executing the neural network model for calculation; and a result output module for outputting the data calculated by the neural network model.

9. A method for predicting the energy requirement of laying hens at different growth stages based on a neural network model, characterized in that: The method comprises using the prediction system as described in claim 7 or 8 to perform prediction and obtain results.

10. The prediction method according to claim 9, characterized in that: The first step is data input, the second step is system calculation, and the third step is output of results.

Citation Information

Patent Citations

  • Method for predicting net energy demand of commercial pigs at different weight stages and application

    CN113361747A

  • Prediction method and device for maintaining net energy demand amount and electronic equipment

    CN114795161A

  • Metabolic energy demand amount dynamic prediction method

    CN116796897A

  • Method for predicting milk yields, tmr nutrient composition to achieve target milk yields, and tmr nutrient composition to achieve target cost, on basis of deep learning-based prediction models

    WO2019216647A1

Cited By

  • Method for rapidly evaluating utilization rate of genetic engineering pig feed

    CN119423008A